All projects
Machine learning · Trading research
Bitcoin price prediction, honestly measured
An LSTM neural network that predicts tomorrow's Bitcoin close from the last ten days, written as a line by line tutorial. It is groundwork for my trading robot, and it shows with real numbers that the model does not beat a simple baseline.
Design challenge
Test an idea properly before trusting it with money, and say so plainly when it fails.
Engineering focus
Seven years of daily candles, scaling, a time based train and test split, sliding windows and a small LSTM in TensorFlow, compared against a naive baseline.
Status
Public learning project on GitHub. It is not a trading tool and gives no financial advice.
Next project
A local Persian intent chatbot